BROOKSPEAK
Crosstalk, Part 1:
The Conversation We Wish Would Stop!
By Douglas Brooks, President
UltraCAD Design, Inc.
Let's talk about crosstalk again, something that is not
well understood by many designers. Consider the trace,
A-B shown in Figure 1. We call that a 'driven' line. Let's
assume that a pulse is traveling down that line from A to
B and is now at the point defined by the 'X'. There is a
nearby trace, C-D, which we will call the 'victim' trace. A
signal may couple between the driven and victim traces at
the point 'X'.
Figure 1
A signal on the driven line, A-B, at point X couples a
noise signal on the "victim" line, C-D, at the same
point.
Now let's think about what is intuitive. There is
likely some capacitive coupling between the two traces.
It may be small, but there is likely to be some. Since
parallel traces sort of look like a transformer, there is
probably some inductive coupling between the traces,
also. In most PCB applications, however, the material
between the traces is probably a very good insulator, so
there is probably no (purely) resistive coupling between
the traces. Since capacitive and inductive effects reduce
with distance, any coupling probably reduces as the
separation between the traces increases. This leads to:
General Rule #1: If we are concerned about
crosstalk, greater separation between traces is better.
If a signal along trace A-B is going to couple noise
into trace C-D, we must be looking at an AC phe-
nomenon. That is, a steady state DC signal on one trace
will not couple into the other one. But if an AC signal
will, then, intuitively, the coupling will be greater (or
This two-column series appeared in Printed Circuit Design, a Miller Freeman publication, November and December, 1997
 1997 Miller Freeman, Inc.
more efficient) the higher the frequency (or frequency
components of the harmonics of the underlying signal.)
More about the specific effects of frequency (and rise
time) next month. But this observation leads to:
General Rule #2: If we are concerned about crosstalk,
lower frequency harmonics and slower rise times are
better.
The effects of coupling can be summarized as fol-
lows:
Mutual Capacitive Coupling--a signal, SC, caused by
capacitive coupling between the two traces, which travels
along the victim trace in both the forward and backward
direction with the same polarity.
Mutual Inductive Coupling--a signal, SL, caused by
inductive coupling between the two traces, which travels
along the victim trace in both the forward and backward
direction with opposite polarity.
Directionality--Crosstalk goes in both the forward
and backward direction. Mutual capacitive and inductive
forward crosstalk are approximately equal and opposite
and tend to cancel. They are approximately equal and
reinforcing in the reverse direction, and therefore tend to
be additive.
Magnitude--Forward crosstalk tends to look like the
driven signal, and (at least in theory) continues to grow
larger the longer is the coupled length A-B (and C-D).
Reverse crosstalk tends to have a rectangular shape (in
response to a step function) that reaches a maximum and
then does not increase further regardless of coupled
length.
Environment--If the two traces are contained within a
homogeneous material (as they are in a stripline environ-
ment), the inductive and capacitive forward crosstalk
components are almost exactly equal and cancel. There-
fore, we generally don't worry about forward crosstalk in
stripline environments. If the surrounding material is NOT
homogeneous (as it is not in a microstrip environment),
the inductive component tends to be larger than the capac-
itive component.
General Rule #3: Therefore, if we are concerned
about crosstalk, keep the sensitive traces in stripline envi-
ronments. (Thus ends the discussion on forward
crosstalk!)
 1997 UltraCAD Design, Inc.
 Figure 2 helps us understand why the reverse
crosstalk waveform tends to be longer than the driven
pulse. Consider a simple signal as it travels down a trace.
We will take a "snapshot" of it at seven points in time,
labeled t=0 through t=6. In order to simplify the analysis,
we will assume the signal is a pulse with zero width, even
though in Figure 2 it is shown with a finite width.
At time t=0 the pulse just starts down the trace. The
reverse crosstalk signal in the victim line reflects off C
with a negative reflection coefficient (Footnote 1) and
begins traveling down the victim line towards D. At time
t=1 the driven pulse causes a coupled signal in C-D to
travel in the reverse direction towards C, while the
coupled pulse from time t=0 has traveled to position t=1.
Skipping to time t=4, the driven signal couples into
C-D at position t=4, but the pulses coupled at times t=3
and t=2 are moving backwards (reverse direction) to-
wards C. Pulses at times t=0 and t=1 have reflected off
the end at C and are now moving with a negative polarity
towards D. In this manner it can be seen that the driven
signal creates a coupled signal in the victim line that (1)
travels in a reverse direction (towards C) until it reflects
off C and then travels the length of the line towards D.
Now, instead of thinking about snapshots in time,
consider the signal in real time as it travels from A to B.
The "front edge" of the signal at each increment of its
travel will couple a noise signal in C-D that will travel
towards C, reflect off C, and travel towards D. The
coupled signal will have a magnitude (to be discussed
next week) and a width equal to the propagation time
from D to C and then back to D (twice the propagation
time of the coupled length.)
The above analysis assumes that the reverse coupled
signal travels towards C and then reflects towards D. But
what if the end at C is terminated in the characteristic
impedance of the trace? In that case it is tempting to say
that there would be no reflection of the reverse crosstalk
waveform. Then, if points B and D were also terminated
in their characteristic impedance, there would be no
reflections at that end, either. Does this mean that we
wouldn't have to be concerned about reverse crosstalk. Figure 2:
And, further, if we are in a stripline environment so that Backwards crosstalk as the driven signal propagates
we don't have to worry about forward crosstalk, would down the trace.
this mean that the entire crosstalk problem would go
away? Never say "never", but next month we will talk
about how to calculate crosstalk, taking terminations into
account.

Footnote 1. Remember that the reflection coefficient is (Rs-Zo)/(Rs+Zo) where Rs is the source resistance at C and
Zo is the characteristic impedance of the trace. Since Rs is normally much less than Zo, the reflection coefficient can
approach -1.
 BROOKSPEAK
Crosstalk, Part 2:
How Loud Is Your Crosstalk?
By Douglas Brooks, President
UltraCAD Design, Inc.

D 1
Last month's column focused on what crosstalk is and
attempted to provide an intuitive understanding of the 1 + ( D / H )2
principles involved. It also concluded that forward Trace 1 Trace 2
Eq. 1a
H H
crosstalk is usually not a problem in stripline environ- Plane
ments. This month I will provide a means of
(approximately) quantifying backwards crosstalk. Figure 1 H2
Estimating crosstalk can be difficult. The ap- Typical crosstalk configuration H 2 + D2
proach shown in this column depends on several sim-
plifying assumptions, and will lead to results that are Eq. 1b
close, but not necessarily precise. The reader would be
wise to calibrate this technique by applying it to
boards that have been designed in the past and are Trace 1 D
known to be either "good" or "bad" with respect to H1 Trace 2
H2 1
crosstalk performance. That will provide insight into
how future boards will perform, based on the results of
Plane
1 + (D / H1* H 2)
2

these calculations. Figure 2 Eq. 2

Trace heights are different
Three concepts are involved in the measurement
approach:

1. Electromagnetic coupling between two traces.

Plane
2. The parallel distance over which that coupling H1a
H2a
occurs. Trace 1 D
3. The effectiveness of any terminations, if they exist. H1b Trace 2 Hna * Hnb
H2b

Electromagnetic coupling:
Plane
Hna + Hnb
Howard Johnson shows in his book “High Speed Figure 3 Eq. 3
Digital Design: A Handbook of Black Magic” (page Use parallel combination to calculate H in
192) that the crosstalk coupling coefficient in parallel stripline environments
traces is proportional to, and does not exceed, Equa-
tion 1a (see Figure 1.) Equation 1a is re-expressed in
Equation 1b. It is clear from Equation 1b what we
already intuitively know: to minimize crosstalk, you Coupling Distance:
minimize H and maximize D. Backwards crosstalk starts out small and increases as
2
If the parallel traces are at different heights, the H the length of parallel coupling increases --- to a point. It
term really becomes the product of the two heights, as reaches a maximum magnitude and then stays constant at the
shown in Figure 2 and Equation 2. D becomes the so-called "critical length"; the length where the round trip
direct distance between the centerline of the traces. propagation time (T ) equals the rise time (T ) of the driving
RT R

Finally, if the traces are between two planes, as signal. So for short traces (T /T =< 1) crosstalk is propor-
RT R

shown in Figure 3, find each H by using a parallel tional to the ratio of T /T . We calculate T with Equation
RT R RT

combination of the heights to each plane, as shown in 4. For long traces, (T /T > 1) calculate crosstalk indepen-
RT R

Equation 3. dent of length.

 TRT = 1.017 aεr + b * Parallel Length * 2 Example:
As a final example, suppose we have the configura-
Where:
εr= Relative dielectric coefficient tion shown in Figure 4. Let εr be 4.5, the rise time of the
a = 1, b = 0 for Stripline coupled pulse be 1 nsec, the parallel length be 10 in. and
a = .475, b = .67 for Microstrip no terminations. The calculated coupling coefficient
would be .05195 (-26 db).
Equation 4
Plane
Use this equation to calculate T RT 6
12
Trace 1

12 Trace 2
6
Plane
Terminations: 16
If the coupled trace is perfectly terminated at the near end
with the characteristic impedance of the trace (Zo), there will Figure 4
be no reflection at the near end and no backwards crosstalk Example
reflecting back down the line. Correspondingly, if the driven
trace is perfectly terminated at the far end, there will be no Calculator:
reflection at the far end that will induce crosstalk in the victim UltraCAD has written a Windows software calcula-
trace. Thus, if there are terminations at both ends of the tor for calculating the crosstalk coefficient. It is dis-
traces, crosstalk will be reduced, and will be proportional to tributed as freeware and can be downloaded from our
the reflection coefficient (measured at the end with the worst web site (www.ultracad.com --- follow the link to tech-
mismatch): nical papers). Download the file ultra_ct.zip, and unzip
( RL − Z 0 ) it in its own directory.
ρ=
( RL + Z 0 )

Assume, for example, the tolerance for Zo is 10% and the

tolerance for R is 3%. The worst case reflection coefficient,
L

then, would be:

(.9 7 − 1.1 )
ρ = = .0 6 3
( 1.1 + .9 7 )

Here is a simplifying trick. For small tolerances like this,

the worst case reflection coefficient can be approximated
with [Tol(Zo) + Tol(RL)]/2. Using the illustration above, this
would become
[10 + 3]/2 =.065

Result: Figure 5
The final result, now, is that the backwards crosstalk UltraCAD’s Crosstalk Calculator
coupling coefficient can be approximated as:

1 Tol ( Zo ) + Tol ( R L ) Acknowledgment: We are indebted to Dr. Howard

( T / T )
1 + ( D / H )2 RT R
2 Johnson (howiej@sigcon.com) for some of the concepts
presented in this column. He was kind enough to spend an
Where the T /T term cannot exceed 1.0, and the final
RT R
hour with my staff recently, educating us on crosstalk. Dr.
term applies only if there are terminations at both ends of the Howard Johnson is president of Signal Consulting, Inc.
traces. (www.sigcon.com), a high-technology consulting firm that
People sometimes like to express coupling in decibels specializes in solving high-speed digital design problems.
(db's). To convert this coupling coefficient to db's, take the He is the author of "High-Speed Digital Design: A
common log (log to base 10) of the coupling coefficient and Handbook of Black Magic" (Prentice-Hall, 1993) and a
multiply it by 20. frequent guest lecturer at the University of Oxford.

crosstalk.pub